bartowski's picture
Update metadata with huggingface_hub
016fa33 verified
metadata
base_model: ibm-granite/granite-20b-code-instruct
datasets:
  - bigcode/commitpackft
  - TIGER-Lab/MathInstruct
  - meta-math/MetaMathQA
  - glaiveai/glaive-code-assistant-v3
  - glaive-function-calling-v2
  - bugdaryan/sql-create-context-instruction
  - garage-bAInd/Open-Platypus
  - nvidia/HelpSteer
library_name: transformers
license: apache-2.0
metrics:
  - code_eval
pipeline_tag: text-generation
tags:
  - code
  - granite
quantized_by: bartowski
inference: true
model-index:
  - name: granite-20b-code-instruct
    results:
      - task:
          type: text-generation
        dataset:
          name: HumanEvalSynthesis(Python)
          type: bigcode/humanevalpack
        metrics:
          - type: pass@1
            value: 60.4
            name: pass@1
          - type: pass@1
            value: 53.7
            name: pass@1
          - type: pass@1
            value: 58.5
            name: pass@1
          - type: pass@1
            value: 42.1
            name: pass@1
          - type: pass@1
            value: 45.7
            name: pass@1
          - type: pass@1
            value: 42.7
            name: pass@1
          - type: pass@1
            value: 44.5
            name: pass@1
          - type: pass@1
            value: 42.7
            name: pass@1
          - type: pass@1
            value: 49.4
            name: pass@1
          - type: pass@1
            value: 32.3
            name: pass@1
          - type: pass@1
            value: 42.1
            name: pass@1
          - type: pass@1
            value: 18.3
            name: pass@1
          - type: pass@1
            value: 43.9
            name: pass@1
          - type: pass@1
            value: 43.9
            name: pass@1
          - type: pass@1
            value: 45.7
            name: pass@1
          - type: pass@1
            value: 41.5
            name: pass@1
          - type: pass@1
            value: 41.5
            name: pass@1
          - type: pass@1
            value: 29.9
            name: pass@1

Llamacpp imatrix Quantizations of granite-20b-code-instruct

Using llama.cpp release b3634 for quantization.

Original model: https://huggingface.co/ibm-granite/granite-20b-code-instruct

All quants made using imatrix option with dataset from here

Run them in LM Studio

Prompt format

System:
{system_prompt}

Question:
{prompt}

Answer:


Answer:

What's new:

New model update

Download a file (not the whole branch) from below:

Filename Quant type File Size Split Description
granite-20b-code-instruct-f16.gguf f16 40.24GB false Full F16 weights.
granite-20b-code-instruct-Q8_0.gguf Q8_0 21.48GB false Extremely high quality, generally unneeded but max available quant.
granite-20b-code-instruct-Q6_K_L.gguf Q6_K_L 16.71GB false Uses Q8_0 for embed and output weights. Very high quality, near perfect, recommended.
granite-20b-code-instruct-Q6_K.gguf Q6_K 16.63GB false Very high quality, near perfect, recommended.
granite-20b-code-instruct-Q5_K_L.gguf Q5_K_L 14.88GB false Uses Q8_0 for embed and output weights. High quality, recommended.
granite-20b-code-instruct-Q5_K_M.gguf Q5_K_M 14.81GB false High quality, recommended.
granite-20b-code-instruct-Q5_K_S.gguf Q5_K_S 14.02GB false High quality, recommended.
granite-20b-code-instruct-Q4_K_L.gguf Q4_K_L 12.89GB false Uses Q8_0 for embed and output weights. Good quality, recommended.
granite-20b-code-instruct-Q4_K_M.gguf Q4_K_M 12.82GB false Good quality, default size for must use cases, recommended.
granite-20b-code-instruct-Q3_K_XL.gguf Q3_K_XL 11.81GB false Uses Q8_0 for embed and output weights. Lower quality but usable, good for low RAM availability.
granite-20b-code-instruct-Q3_K_L.gguf Q3_K_L 11.74GB false Lower quality but usable, good for low RAM availability.
granite-20b-code-instruct-Q4_K_S.gguf Q4_K_S 11.67GB false Slightly lower quality with more space savings, recommended.
granite-20b-code-instruct-Q4_0.gguf Q4_0 11.61GB false Legacy format, generally not worth using over similarly sized formats
granite-20b-code-instruct-Q4_0_8_8.gguf Q4_0_8_8 11.55GB false Optimized for ARM inference. Requires 'sve' support (see link below).
granite-20b-code-instruct-Q4_0_4_8.gguf Q4_0_4_8 11.55GB false Optimized for ARM inference. Requires 'i8mm' support (see link below).
granite-20b-code-instruct-Q4_0_4_4.gguf Q4_0_4_4 11.55GB false Optimized for ARM inference. Should work well on all ARM chips, pick this if you're unsure.
granite-20b-code-instruct-IQ4_XS.gguf IQ4_XS 10.94GB false Decent quality, smaller than Q4_K_S with similar performance, recommended.
granite-20b-code-instruct-Q3_K_M.gguf Q3_K_M 10.57GB false Low quality.
granite-20b-code-instruct-IQ3_M.gguf IQ3_M 9.59GB false Medium-low quality, new method with decent performance comparable to Q3_K_M.
granite-20b-code-instruct-Q3_K_S.gguf Q3_K_S 8.93GB false Low quality, not recommended.
granite-20b-code-instruct-IQ3_XS.gguf IQ3_XS 8.66GB false Lower quality, new method with decent performance, slightly better than Q3_K_S.
granite-20b-code-instruct-Q2_K_L.gguf Q2_K_L 8.00GB false Uses Q8_0 for embed and output weights. Very low quality but surprisingly usable.
granite-20b-code-instruct-Q2_K.gguf Q2_K 7.93GB false Very low quality but surprisingly usable.
granite-20b-code-instruct-IQ2_M.gguf IQ2_M 7.05GB false Relatively low quality, uses SOTA techniques to be surprisingly usable.

Q4_0_X_X

If you're using an ARM chip, the Q4_0_X_X quants will have a substantial speedup. Check out Q4_0_4_4 speed comparisons on the original pull request

To check which one would work best for your ARM chip, you can check AArch64 SoC features(thanks EloyOn!).

Embed/output weights

Some of these quants (Q3_K_XL, Q4_K_L etc) are the standard quantization method with the embeddings and output weights quantized to Q8_0 instead of what they would normally default to.

Some say that this improves the quality, others don't notice any difference. If you use these models PLEASE COMMENT with your findings. I would like feedback that these are actually used and useful so I don't keep uploading quants no one is using.

Thanks!

Credits

Thank you kalomaze and Dampf for assistance in creating the imatrix calibration dataset

Thank you ZeroWw for the inspiration to experiment with embed/output

Downloading using huggingface-cli

First, make sure you have hugginface-cli installed:

pip install -U "huggingface_hub[cli]"

Then, you can target the specific file you want:

huggingface-cli download bartowski/granite-20b-code-instruct-GGUF --include "granite-20b-code-instruct-Q4_K_M.gguf" --local-dir ./

If the model is bigger than 50GB, it will have been split into multiple files. In order to download them all to a local folder, run:

huggingface-cli download bartowski/granite-20b-code-instruct-GGUF --include "granite-20b-code-instruct-Q8_0/*" --local-dir ./

You can either specify a new local-dir (granite-20b-code-instruct-Q8_0) or download them all in place (./)

Which file should I choose?

A great write up with charts showing various performances is provided by Artefact2 here

The first thing to figure out is how big a model you can run. To do this, you'll need to figure out how much RAM and/or VRAM you have.

If you want your model running as FAST as possible, you'll want to fit the whole thing on your GPU's VRAM. Aim for a quant with a file size 1-2GB smaller than your GPU's total VRAM.

If you want the absolute maximum quality, add both your system RAM and your GPU's VRAM together, then similarly grab a quant with a file size 1-2GB Smaller than that total.

Next, you'll need to decide if you want to use an 'I-quant' or a 'K-quant'.

If you don't want to think too much, grab one of the K-quants. These are in format 'QX_K_X', like Q5_K_M.

If you want to get more into the weeds, you can check out this extremely useful feature chart:

llama.cpp feature matrix

But basically, if you're aiming for below Q4, and you're running cuBLAS (Nvidia) or rocBLAS (AMD), you should look towards the I-quants. These are in format IQX_X, like IQ3_M. These are newer and offer better performance for their size.

These I-quants can also be used on CPU and Apple Metal, but will be slower than their K-quant equivalent, so speed vs performance is a tradeoff you'll have to decide.

The I-quants are not compatible with Vulcan, which is also AMD, so if you have an AMD card double check if you're using the rocBLAS build or the Vulcan build. At the time of writing this, LM Studio has a preview with ROCm support, and other inference engines have specific builds for ROCm.

Want to support my work? Visit my ko-fi page here: https://ko-fi.com/bartowski